里约"国产"大模型被曝系拼接产物
推荐指数 50.0 NO. 021 · 2026.06.15
发布2026/06/14Score165Comments91
为什么值得看
巴西里约热内卢政府高调推出的397B参数"自研"LLM Rio-3.5-Open，被社区发现实为Nex-N2_pro与Qwen的加权合并模型（6:4比例），非独立训练。此事暴露地方政府AI政绩工程的典型造假路径，也给依赖"国产自研"叙事的采购方敲响警钟。
编辑判断
模型合并（model merging）本身是完全合法的技术手段，Mistral、Sakana AI 等都做过，问题出在里约政府把 merge 包装成"homegrown"自主训练。这种操作在发展中国家 AI 采购中可能形成模板：用开源权重加轻微微调，换取巨额财政补贴和民族情绪溢价。
对 AI 工程师的实际影响是：以后评估"某国自研模型"时，必须用 SHAP 值分析、权重相关性检测或逆向 logit 对比来验证独立性，不能只看发布会通稿。社区已经有人在做自动化 merge 检测工具，这类"模型血统"验证可能会成为 MLOps 的新标准环节。
社区反馈
负面 75 条评论
核心争论：模型合并是否算创新，以及政府AI政绩工程造假的典型路径
The municipality of Rio de Janeiro (via its IT company IplanRIO) released Rio-3.5-Open-397B, presented as a homegrown Qwen3.5 fine-tune that beats comparable open models on benchmarks. The linked issue argues it's actually a weighted merge of ~60% Nex-N2 Pro + ~40% Qwen3.5-397B-A17B - Nex-N2 having
So the problem isn’t in the missing attribution to Qwen, but with the fact that they didn’t mention Nex-N2 Pro right?
The problem is that they claimed to have made a big achievement with their home grown post training, and they expected to receive a lot of praise for it. Then researchers looked at the weights and there is no post training at all. They are now attributing both models they merged, but their excuse fo